Sei sulla pagina 1di 30

Chapter 6

Game Theory- Normal Form Games

Key words:

Game theory, strategy, finite game, zero-sum game, pay-off matrix, dominant

strategy, value of the game, fair game, stable solution, saddle point, pure strategy, mixed strategy,

expected pay-off.

Suggested readings:

1. Gupta P.K. and Mohan M. (1987), Operations Research and Statistical Analysis, Sultan

Chand and Sons, Delhi.

2. Hillier F.S. and Lieberman G.J. (2005), Introduction to Operations Research, (8 th edition),

Tata-McGraw Hill Publishing Company Limited.

3. Johnson R.D. and Bernard R.S. (1977), Quantitative Techniques for Business Decisions,

Prentice hall of India Private Limited

4. Raiffa H. and Schlaifer R. (1968), Applied Decision Theory, MIT Press.

5. Swarup K., Gupta P.K. and Mohan M. (2001), Operations Research, Sultan Chand and Sons,

Delhi.

6. Watson J., (2002), Strategy: An Introduction To Game Theory, W.W. Nortan & Company

7. Osborne M.J., (2001) An Introduction to Game Theory, Oxford University Press.

179

6.1

Introduction

The maximization of expected value criteria, which we have been discussing till now, is an efficient

criterion in the sense that depending upon all the available information, we have to choose a strategy

among all possible alternatives, so that we receive the maximum possible benefit. Here we assume that

the outcome of a decision is a random variable with some well-defined probability distribution.

In

other words we say that the outcome of decision is determined by some neutral factor (viz. nature).

Obviously, this neutral factor does not have any interest in the benefits or losses, which we are

receiving as a participant and hence is not an active participant in the process.

However, the situation may not always be so simple.

Sometimes, the outcome of a decision is not

controlled by a neutral factor but by a well-informed and intelligent adversary who has an active

interest in the process. This is the situation of conflict (and competition).

The world is full of conflicting situations. In fact the resources in the world are limited and when one

party tries to increase its share in the available resources, it does so at the cost of any other party. This

is the situation of conflict.

Have a look at the world surrounding you and you will find conflicts

everywhere.

Labour-management

relationship,

political

and

military

conflicts,

competitions,

maneuvers, marketing and advertising tactics, these all are different faces of conflicts. In all these and

many more situations, one party tries to maximize its benefits at the cost of others.

While resolving these disputes is a time-consuming and complex job, it is possible to develop optimal

strategies mathematically for such conflicts. Off course, when we are developing strategies, we make

some simplifying assumptions, (which we know may not always be true).

6.2 Game Theory

The techniques of developing optimal strategies for dealing with conflicting and competitive situations

(whenever these conflicts can be expressed in mathematical terms) have been termed as game theory.

We define some terms associated with the game theory

180

(i)

Strategy

A strategy is a comprehensive plan of action, formulated by a player (an interested

and active party in the game), who is well informed of all the alternatives available to him and to his

adversary (competing player).

A strategy can be good or bad.

possibilities.

The only requirement is that it should be complete and cover all the

(ii) Finite game

When the total number of possible strategies in a game is finite, it is called a

finite game. In the other situation, the game is an infinite game.

(iii) Zero-sum game

Zero-sum games are those games in which one player gains exactly the

same amount, which the other player(s) loose so that their net gains is equal to zero.

(iv) Non zero-sum game

Zero-sum games are those games in which gain of one player is not

necessarily equal to the loss of the other or vice versa.

(v) Pay-off (game) matrix

A pay-off matrix is a tabular representation of the pay-offs of one

competitor, which are associated with his strategies in response to the strategies of the other player.

Consider two players A and B playing a zero-sum game. Let A has m strategies numbered A 1 , A 2 A m

available to him and B has n strategies numbered B 1 , B 2

B n , available to him.

Let the gain of A,

when he chooses i th strategy in response to the j th strategy chosen by B be given by g ij . Then the pay-

offs of A can be represented as follows:

G

=

A

1

A

2

M

A

m

B

1

g

g

M

g

11

21

m

1

B

B

2

g

g

M

g

12

22

m

2

B

n

M

g

1 n

g

2 n

g

mn

⎟ ⎠

The matrix G is called the pay-off matrix of player A. If g ij > 0, A has gained and if g ij < 0, then A has

lost an amount g ij .

Since the game is a zero-sum game, so whatever is the gain of A is loss of B.

181

Assumptions of gaming problems

Game theory is meant for developing a rational criterion

for choosing a strategy among several possible strategies. For developing such criteria, we make some

assumptions:

(i)

The number of players in the game is (in general) finite.

(ii)

The interests of the players clash and each player is choosing his strategy solely for his

welfare.

(iii)

Each player is well aware of all the strategies available to him and to his opponents.

(iv)

All the players are making their moves simultaneously, without knowing the choices,

which the other players have made.

(v)

The outcome of the game depends upon the moves made by different players; and

(vi)

All the players are rational players.

6.3 Solving a zero-sum game

In general, the games are zero-sum games.

For the simplicity of presentation, we assume that the

games are two- players’ games. We define some terms associated with the solution of the games.

(i) Dominant strategy

Consider the following game (G 1 )

B

A

1

A

2

B

1

7

5

B

2

4

2

B

3

6

4

This game matrix suggests that the two players A and B are playing a game with A having two (viz. A 1

and A 2 ) and B having three (viz. B 1 , B 2 and B 3 ) strategies.

In this case A would always opt for the

strategy A 1, as it would yield him better pay-offs than the pay-offs yielded by the strategy A 2 . We say

that A 1 is a dominant strategy.

A strategy is said to be a dominant strategy if it always yields better (or at least equal) pay-offs than the

other strategies irrespective of the strategies opted by the other player(s), i.e., superior strategies

(resulting in higher pay-offs) dominate the inferior ones (resulting in lower pay-offs).

situations, inferior strategies can always be strike off.

182

In such

Consider the following example

Example 1:

Two firms, ABC Ltd. and XYZ Corp. are competitors in the market of electronic

goods.

In order to increase its market share, each of the firm can opt any of the following three

strategies: high advertising, moderate advertising or low advertising.

Corresponding to different

possible conditions, the pay-offs in terms of percent market share are given below:

High(1)

XYZ Corp.

Moderate(2)

Low(3)

ABC Ltd.

High(1)

2

3

5

Moderate(2)

2

0

6

Low(3)

0

2

-1

The managements of the two firms are interested in determining the optimal strategies.

Sol:

For XYZ corp., as such there is no dominant strategy.

But for ABC Ltd, strategy 1 is

dominant over strategy 3. We eliminate the dominated strategy and the reduced pay-off matrix is given

by

High(1)

XYZ Corp.

Moderate(2)

Low(3)

ABC Ltd.

High(1)

2

3

5

Moderate(2)

2

0

6

In this reduced matrix, XYZ corp. would try to minimize its losses so it would eliminate those

strategies that are paying a higher pay-off to ABC Ltd. and hence strategies 1 and 2 for XYZ corp.

becomes the dominant over strategy 3 and hence strategy 3 can be eliminated.

we have

183

After this elimination

ABC Ltd.

High(1)

XYZ Corp.

Moderate(2)

High(1)

2

3

Moderate(2)

2

0

At this point, ABC Ltd. will again try to maximize its gains and for that it would eliminate strategy 2 so

we have

ABC Ltd.

High(1)

XYZ Corp.

Moderate(2)

High(1)

2

3

Finally, XYZ Corp. would settle at strategy 1, which is minimizing the pay-off to ABC Ltd. Hence the

optimal strategy for both the players would be to go for high advertising.

With the selection of the optimal strategies, the market share of ABC Ltd would increase by 2%. This

is the value of the game.

We define the following terms

Value of the game

The pay-off received by the player (whose pay-off matrix is given) when

both the players play optimally, is called the value of the game.

Fair game

A game that has a value 0, i.e., neither player is neither a loser nor a winner, is called a

fair game.

The above game is not a fair game.

In the game G 1 , A will always choose the strategy A 1 .

But what would be the strategy chosen by B?

Since whatever is the gain of A, it is the loss of B so naturally B would try to minimize his loss (or gain

of A). Then B’s obvious choice will be the strategy B 2 . The value of the game in this case is 4.

(ii) The maximin and minimax principle

Consider the following game:

184

B

A

1

A

2

A

3

B

1

1

2

0

B

2

3

1

2

B

3

6

-3

1

If A chooses A 1 , then B will choose B 1 so that the gain of A (or loss of B) is minimized (to 1).

For

strategy A 2 chosen by A, B will select the strategy B 3 . Similarly for strategy A 3 of A, B will choose the

strategy B 1 . Thus the minimum pay-offs assured to A are

Table 6.1

Strategy

Minimum pay-off

A

1

1

A

2

-3

A

3

0

Among these minimum assured pay-offs A would try to maximize his pay-off. Thus he would choose

the strategy A 1 that is the maximin (maximum among minimum) strategy.

Similarly, the player B is playing to minimize his losses.

corresponding to different strategies selected by him:

We have the following loss table

Table 6.2

Strategy

Minimum pay-off

B

1

2

B

2

3

B

3

6

Then the strategy B 1 , which corresponds to the minimum loss among the maximum possible losses, is

the minimax strategy.

Thus minimax-maximin principle is that rule on the basis of which one player tries to minimize the

worst possible losses and the other player tries to maximize the minimum assured gains.

185

Example 2:

For the firms ABC Ltd. and XYZ Corp., let the pay-off matrix be given by

High(1)

XYZ Corp.

Moderate(2)

Low(3)

Row minimums

ABC Ltd.

High(1)

Moderate(2)

Low(3)

Column maximums

-4

-1

7

-4

2

0

3

0

Maximin value

3

-2

1

-2

3

0

7

Mini max value

On solving this game using dominance principle, we have

Step 1:

XYZ Corp.

 

High(1)

Moderate(2)

 

High(1)

-4

-1

 

ABC Ltd.

Moderate(2)

2

0

 

Low(3)

3

-2

Step 2:

 

XYZ Corp.

High(1)

Moderate(2)

 

ABC Ltd.

Moderate(2)

2

0

 

Low(3)

3

-2

Step 3:

 

XYZ Corp.

 

Moderate(2)

 

ABC Ltd.

0 -2
0
-2

Moderate(2)

Low(3)

 

186

Step 4:

ABC Ltd.

XYZ Corp.

Moderate(2) Moderate(2) 0
Moderate(2)
Moderate(2)
0

Thus the value of this game is 0. In this game maximin gain of ABC Ltd. is same as the minimax loss

of XYZ Corp. Such a game is said to possess a saddle point.

Saddle point

A saddle point of a game (if it exists) is that point in the pay-off matrix (of player A)

where maximin gain of A is equal to the minimax loss of player B. In such a case, the saddle point is

the value of the game.

To find a saddle point

(i)

Find the minimum values in the rows;

(ii)

Mark the maximum of these minimums. This is the maximin value (or gain) for player A;

(iii)

Find the maximum values in columns;

(iv)

Mark the minimum of these maximums. This is the minimax value (or loss) for player B;

(v)

If the maximin gain of A = the minimax loss of B, the point in the pay-off matrix is the

saddle point.

Stable (equilibrium) solution

If a saddle point exists in a game, the corresponding pair of

strategies is a stable (equilibrium) solution.

The solution is stable due to the fact that in presence of a saddle point neither player is in a position to

take advantage of the information about the opponent’s choice. In order to optimize his pay-off, each of

the players has to stick to that strategy, which would lead him to the saddle point. Hence the solution is

a stable solution.

(iii) Use of mixed strategies

Consider the following game

187

 

Player

B

 

1

2

3

Row minimums

 

1

0

-2

2

-2 Maximin value

Player

A

2

5

4

-3

-3

 

3

2

3

-4

-4

Column maximums

5

4

2

 

Minimax value

In this game, since minimax value is not equal to maximin value so the game does not have a saddle

point. Since both the players are rational players so they are choosing their strategies in such a manner

as to maximize their benefits.

Starting with A, the player is assured of minimum gain of two units.

And subsequently he will choose strategy 3. Anticipating this situation, player B will try to minimize

the gains of player A and hence he will choose that strategy 3.

Now player A, being aware of the

intentions of B, will choose strategy 2 to maximize his pay-off. In this position, B will have to go for

strategy 3 and the whole process will repeat itself. Thus this game does not have a stable solution.

In such games the fact persists that on having information about the opponent’s moves; each player can

improve his position. Then, in such situations, in order to arrive at a solution, it is necessary that none

of the players should have any advance knowledge about the opponent’s moves. So, in place of having

some criterion about choosing a single strategy to be used definitely, a probability distribution is used

to choose among various possible (and acceptable) strategies. By doing so, none of the players will, in

advance, be able to know about the strategies to be used by his opponent or to be used by himself.

We define the following terms:

Pure strategy

When a player knows with certainty the strategy that he is going to use, the strategy

is called a pure strategy. In fact, a pure strategy is a strategy actually used.

Mixed strategy

When a player does not know about the strategies to be used with certainty, but uses

a probability distribution to determine it, then that probability distribution is called a mixed strategy.

188

Consider a game, being played by two players A and B that have no saddle point. Let player A has m

strategies available to him and the player B has n strategies available to him. Define

p

i

q

j

=

=

P

P

(player

(player

A

B

uses

uses

i

j

th

th

strategy) ,

strategy) ,

i

j

=

=

1 , 2

1 , 2

m

n

.

;

For player A, the pure strategies are (1, 2

m) and the mixed strategies are (

p

1

,

p

2

p

m )

Expected pay-off: A measure of performance of mixed strategies

Let g ij be the pay-off to player

A when the player B uses the strategy j and he uses the strategy i, i = 1,2… m and j = 1,2…n. Then the

expected pay-off of the player A is defined as

(

E G

)

m

n

= ∑∑

i

=

1

j

=

1

g

ij

p q

i

j

1 1

, 0 and that used by the player B are

Suppose that the mixed strategies used by the player A are

0 ,

1

2

,

1

2

. Then the expected pay-off to player A is,

⎝ ⎜

2

, 2

 

Player

2

B

1

 

3

1

1(0)

⎝ ⎜

2

⎠ ⎟

⎝ ⎜

2

⎠ ⎟

 
 

2

 

0

-2

2

 
 

Player

A

5

4

-3

 

2

3

-4

 

(

E G

)

=

1 ⎞⎛ 1

2 ⎠⎝ 2

⎟⎜

(

224

− +

+

3

)

=

1

4

The expected pay-off can be used to choose that strategy which will maximize the minimum

guaranteed pay-off (the maximin criterion).

Let the maximin pay-off be denoted by ν .

189

The optimal strategy for player B will minimize the maximum expected loss. Let this value be denoted

by ν . In the games, which have no saddle point, if only pure strategies are used, then

ν ν

Thus the players keep on flipping their strategies in a hope to improve their positions.

For a game to be a stable game, it is necessary that

ν =ν .

We state the following theorem

Theorem 6.1

If the mixed strategies are allowed, then the optimal pair of mixed strategies

(according to the maximin criterion) will correspond to a stable solution and in that case ν

= ν = ν.

in this case, ν will be the value of the game.

6.4 Calculus approach to solve a 2X2 game

In this method, the expected value of the game for a player is maximized and the probability (mixed

strategy) corresponding to which this maximum value exists, is obtained.

Consider the following game

Player

A

B

1

Player

B

B

2

A

1

A

2

a

a

11

21

a

12

a

22

To find the optimal mixed strategy for this game, we have to optimize expected gain to any one player.

Suppose we want to maximize the expected gain to player A. We assign probabilities x and 1-x to A of

choosing strategy A 1 and A 2 respectively. Similarly probabilities y and 1-y are assigned to player B for

choosing strategies B 1 and B 2 respectively. Then expected gain to player A is given by

E(A)

=

xya

11

+ x(1y)a

190

12

+ (1x)ya

21

+ (1x)(1y)a

22

To determine optimal values of x and y, we differentiate this function partially with respect to x and y at

0,0, and we get

The two equations yield

 

E A

(

)

x

 

=

and

E A

(

)

=

y

(

y a

11

+

a

ya

xa

11

11

22

+

+

(1

(1

y ) a

12

x ) a

21

a

12

a

21

)

=

ya

21

xa

12

a

22

(1

(1

a

12

y ) a

22

x ) a

22

y

x ( a

x

 

a

22

a

12

 

a

11

+

a

22

a

12

a

21

 

a

22

a

12

a

21

)

=

a

 

a

22

a

21

a

11

+

a

22

)

y a

12

+

(1

a

12

a

)

x ya

21

21

+

=

11

=

+

and

Substituting these values of x and y in E (A), we have

E

(

A

)

=

xya

11

+

x

(1

=

a

11

a

22

a

12

a

21

22

(1

a

21

x

)(1

y a

)

22

=

=

0

0

a + a − a − a 11 22 12 21 For second order conditions,
a
+
a
a
a
11
22
12
21
For second order conditions, we have
2
2
E
( A
)
E
(
A
)
=
0;
=
0;
and
2
2
∂ x
∂ y
2
E (
A
)
=
a
+
a
a
a
11
22
21
12
∂∂ x
y
2
2
2
2
⎛∂
E
( A
)
⎞ ⎛∂
E
(
A
)
⎞⎛∂
E
(
A
)
>
0
⎟⎜
2
2
∂∂ y x
∂ x
∂ y
⎠⎝

Thus expected gain of player A for these values of x and y is maximum.

We illustrate the results with the help of the following examples.

191

Example 3:

Let the pay-off matrix of two players A and B is given as follows:

Player B

 

y

1-

y

Player A

x

1

7

 

1-

x

5

2

Player A selects his strategies with probabilities x and 1-x respectively and the player B chooses his

strategies with probabilities y and 1-y respectively. Find the solution to the game.

Sol:

Expected gain of player A is

E( A) = xy + 7x(1y) + 6(1x) y + 2(1x)(1y)

This expected gain is to be maximized for player A. for that, we differentiate the expression partially

with respect to x and y at zero. We have

And

The value of the game is

E ( A

)

 

6

 

x

=

y

+

7(1

y

)

y

⇒ −

10

y

+

5

=

0

 

 

=

1

=

1

 

y

y

2

E ( A

)

7

 

6(1

 

)

 

=

x

x

+

x

 

y

⇒ −

10

x

+

4

=

0

 

x

=

2

and

 

1

x

=

3

     
 

5

5

E

(

A

)

=

 

xy +

7

x

(1

)

 

=

4.0

 

2(1

2(1

y

)

x

)

y + − x y +

6(1

)

=

=

2(1

0

0

x y

)(1

)

Example 4:

In a coin-tossing game, Player A wins a unit amount when there is a sequence of HH

and wins nothing if there is a sequence of TT. However, a sequence of HT or TH results in a loss of

half unit to player A. Find the best strategies for two players and the value of the game.

192

Sol:

The pay-off matrix of the player A is as given below:

Player A

A

A

2

Player B

BB

1

2

1

1 -0.5

-0.5

0

Since the game does not possess any saddle point, we use mixed strategies for two players. Let Player

A selects his strategies with probabilities x and 1 -x respectively and the player B chooses his strategies

w ith probabilities y and 1-y respectively. Then

 

a

a

0.5

 
 

x 21

22

 

=

   

=

=

 

a

+ a

a

a

1 0.5

+

+

0.5

 

and

=

11

a

22

22

a

12

 

21

=

0.5

=

y 12

a

11

+ a

22

a

12

a

21

1 0.5

+

+

0.5

Then 1-

x

=

1-

y

=

0.75

 

T he expected value of the game is

 
 

=

(0.5)(0.5)

 

= −

0.25

=

0.125

 

υ

 
 

1

+

0.5

+

0.5

 

2

6.5

Graphical solu on procedure

ti

 

Consider again the above pay-off matrix:

 

Player B

 
 

1

2

3

1

0

-2

2

 

Player

A

2

5

4

-3

 

3

2

3

-4

0.25

0.25

Foe player A, strategy 2

is dominant over st

strategies for player A are ( p , p

1

2

); p

2

=

1

p

1

rategy 3, so after eliminating str

and the reduced pay-off matrix is

ategy 3, the mixed

193

Player

A

1(

p

1

)

2(1-

p

1

)

Player B

1(

q

1

)

2

(

q

2

)

3

(

q

3

)

 

0

 

- 2

 

2

 

5

4

 

- 3

 

If player B chooses strategy 1, then the expected pay-off of player B is

Similarly for strategy 2, the expected pay-off is

2

p

1

(

1

p )

1

3

= −

3

+

5

p 1

.

214

p

1

+

(

p )

1

=

4

6

0

p

1

+

(

1

p )

1

5

=

5

5

p .

1

p

1 and that for strategy 3 is

In graphical method of solving a game problem, we plot these expected pay-offs on a graph.

Player B is playing to minimize the expected pay-off of player A, so he will choose that strategy, which

corresponds to the lines lying in the bottom part of the graph, i.e.,

3

+

5

p

1

or

4

6

p

1

.

Now, the player A is playing to maximize this expected pay-off so he chooses that strategy which will

maximize the pay-off yielded from these lines, i.e. the intersection of these two lines which is given by

− +

3

5

p

p

1

1

=

=

7

11

4

6

and

p

1

p

2

194

= 4

Expected pay-off . 6 . 5 5-5p 1 4. 3. 4-6p 1 Maximin point 2.
Expected pay-off
.
6
.
5
5-5p 1
4.
3.
4-6p 1
Maximin point
2.
1.
0.
.
.
.
p 1
.
1
0.25
0.5
0.75
-1.
-3+5p 1
-2.
-3.
-4.
Fig. 6.1
⎛ 7
4
Thus, the optimal strategies for player A are
,
, 0 ⎟ and the value of the game is
11
11
⎛ 7 ⎞
2
− +
3
5
=
⎝ ⎜
11
11

Now, we proceed to find the optimal strategies for player B:

The expected pay-off of player A is

q (

1

5

5

p ) + q (

1

2

4

6

3

p ) + q (− +

1

3

5

p )

1

and by the minimax

theorem as ν

= ν

= ν , so if the optimal strategies for the player B are (

satisfy the maximum expected pay-off to A, I.e.,

q

'

1

(

5

5

p

1

)

+

q

'

2

(

4

6

p

1

)

+

q

0

'

3

(

p

3 +

1

5

1

p

1

)

q' ,q' ,q' )

1

2

3

, these must

ν =ν =

2

11

195

Put p

Since

=

7

1 11

so

20

2

>

11

11

So we have

and q '

1

20

'

+

2

 

'

+

2

'

=

2

11

q

1

11

q

2

11

q

3

11

Also

 

q

'

1

+

q

'

2

+

q

'

3

=

1

q

0

'

2

, so the only possible value for


=

 

q '

1

is

q'

2

0

11

;

p

1

2

7

 

=

11

;

p

1

 

11

(

4

6

p

1

)

+

q

'

3

(

3 +

5

p

1

)

1

=

1

0 ,

As the left hand side is a linear expression so the only possible solution to this relation is

Let p 1 = 0 and 1,

q '

2

(

4

and

6

p

1

)

+ q

'

3

(

4

q

'

2

2

q

'

2

3

q

'

+

2

3

q

− + p

3

5

=

'

3

2

11 2 =
11
2
=

11

1

Solving these two equations, we have

q

'

2

=

5

q

'

=

6

11

3

11

and

Thus the optimal strategies for player B are

⎝ ⎜

0 ,

5

6

11

,

11

⎠ ⎟

.

) =

2

11

6.6 Linear programming approach

The graphical method, although a convenient method of solving a game problem, which does not have

a saddle point, suffers from a major limitation. This method can be used only when one of the players

196

has only two (undominated) pure strategies.

But, this may be too simplified a situation.

In large

games, where each of the players may have several strategies to choose from, this method cannot be

employed conveniently.

In such problems, linear programming approach is adopted.

The expected pay-off to player A is

For a strategy (

p

1

,

p

2

,

,

p

m

) to be optimal,

E ( G

)

m

n

= ∑∑

i

=

1

j

=

1

E(G) =ν =

g

ij

p q

i

ν.

j

Since this relationship holds for each of the opponent’s strategy (

q

1

,

Also,

for

j

=

1 2

,

m

n

n

∑∑

i

=

1

j

=

1

, and putting

m

i = 1

q

j

=

g

ij

p

i

1

&

ν

g

ij

p q

i

j

=

n

m

∑ ∑

q

j

j

=

1

i

=

1

n

=

j

1

q

ν

j

=

g

ij

p

ν

i

q

k

as

q

2

,

=

n

=

j

1

Then the problem can be restated as follows

 

maximise

p

m

+ 1

(

subject to

 

g

11 p

1

g

21

p

2

+

g

12 p

1

g

22

p

2

+

 

M

 

g

1

n

p

1

+ 2

g

n

p

2

+

p

1

+

p

2

+

+

p

m

p

i

0

i

=

+

+

197

=

ν

)

+ g

+ g

m

m

1

2

p

m

p

m

+ g

mn

p

= 1

1 2

,

m .

m

-

-

-

p

m +

1

p

m +

1

p

m

+ 1

,

q

n

0

) so we have,

k

j

q

j

=

1

0

0

0

For player B, the equivalent problem is

minimise

subject to

M

q

m

g

g

11

21

+ 1

q

1

q

1

(

+

+

=

ν

)

g

g

12

22

q

2

q

2

g

q

q

m

1

i

1

q

1

+

q

+

g

2

+

0

m

2

q

+

+

+

2

+

q

i

n

=

Thus the two problems are dual problems.

+

+

g

1

n

q

n

g

2

n

q

m

-

q

n +

1

-

q

n +

1

+

=

1 2

,

g

mn

1

n

q

m

-

q

n +

1

0

0

0

In order to ensure that p m+1 and q n+1 are non-negative, a sufficiently large quantity can be added to it. As

this quantity is added to every entry, the optimal strategy will remain the same.

To demonstrate this approach, we again consider the earlier problem, which using linear programming

approach can be restated as follows:

subject to

maximise

p

3

(

=

ν

)

5

p

2

p

3

2

p

1 +

4

p

2

p

2

p

p

1

1

3

+ p

2

p

2

- p

3

= 1

3

0

0

0

p i

0

i = 1 2

,

.

The following algorithm gives the optimal solution to the problem

198

Simplex method- Two-phase procedure

Phase 1- Iteration 1

Table 6. 3

Basic

p

1

p

2

p

3

Sp 4

Sp 5

Sp 6

Rp 7

Rp 8

Rp 9

Rp 10

Solut

variables

     

ion

z

(min)

1.00

7.00

3.00

-1.00

-1.00

-1.00

0.00

0.00

0.00

0.00

1.00

Rp 7

0.00

5.00

1.00

-1.00

0.00

0.00

1.00

0.00

0.00

0.00

0.00

Rp 8

-2.00

4.00

1.00

0.00

-1.00

0.00

0.00

1.00

0.00

0.00

0.00

Rp 9

2.00

-3.00

1.00

0.00

0.00

-1.00

0.00

0.00

1.00

0.00

0.00

Rp 10

1.00

1.00

0.00

0.00

0.00

0.00

0.00

0.00

0.00

1.00

1.00

Lower

0.00

0.00

-

               

bound

 

Upper

               

bound

Unrestric

n

n

y

               

ted

(y/n)?

Phase 1- Iteration II

 

Table 6. 4

 

Basic

p

1